1. Introduction

Scientific examination of paintings is carried out routinely in major galleries and museums to assist in conservation treatment and as part of technical or art historical examinations. Increasingly emphasis is being placed on non-destructive and non-invasive methods of analysis and preventive conservation. However, to study the paint and varnish layers, it is still current practice to take tiny samples from a painting to mount and examine under a microscope. Since the 1930s, the examination of cross-sections has been essential to identifying the pigments and media, their method of application, including relative thickness, the composition of superimposed layers and signs of deterioration and previous alteration [1]. However, since conservation practice and ethics limit sampling to a minimum (normally less than one square millimetre in size) and to areas along cracks and edges of paintings, which are often unrepresentative of the whole painting, results from such analyses cannot be taken as representative of the painting as a whole. A non-invasive technique for examining the paint and varnish layers would not only reduce the need to take samples from paintings, but also enable examinations on any area of a painting.

Currently, routine non-invasive methods of examination include X-radiography, infrared reflectography, macrophotography, UV-fluorescence and raking light imaging [2]. The last three methods give information mostly on the conditions of the surface of a painting. X-radiography [3] is routinely employed to examine the structure of the support of a painting, as well as details of areas painted with pigments containing heavy elements. Infrared reflectography [4] is one of the most useful techniques for art historians studying the preparatory drawings or underdrawings underneath the painted layers, which would be otherwise invisible to the eye. Both X-radiography and infrared reflectography reduce the 3D information of a painting into 2D, thus losing the detailed information perpendicular to the painting plane.

Laser interferometry such as holography and speckle interferometry have been applied to works of art to detect structural defects [5,6,7,8]. However, since these methods employ a laser source that has high temporal coherence, they do not offer the ability to distinguish between successive reflective layers.

In the context of an EU-funded project, CRISATEL, a pilot study was conducted to evaluate whether ellipsometry can be used to measure varnish thickness and optical properties on paintings. Initial studies found that it was possible to measure thin smooth varnish layers on a smooth glass substrate or a smooth machine painted layer [9]. However, after further tests on more realistic samples, we concluded in collaboration with the leading ellipsometer company SOPRA in Paris (http://www.sopra-sa.com/) that varnish layers on real paintings are too thick and too rough to be measured by any ellipsometer (including those operating in the infrared). Ellipsometry is mainly designed for the semi-conductor industry, to measure ultra-smooth and relatively thin transparent film deposits such as SiO₂ on metal substrates [10,11].

Confocal microscopy has been tried on tiny samples for depth resolved imaging. Theoretically it is a non-destructive and non-invasive technique, but in practice hazardous because of the close working distance (a few mm) required for high depth resolution imaging [12].

Recently a novel technique - Optical Coherence Tomography (OCT) has been successfully applied to art conservation to produce cross-sectional images of paintings and archaeological objects non-invasively [13,14,15]. OCT is basically a fast high-resolution 3-D scanning Michelson interferometer [16]. It was developed specifically for producing high resolution 3-D images of the eye and other biological tissues. OCT uses the low coherence interferometry principle, where the depth resolution is not like in confocal microscopy given by the numerical aperture of the microscope objective, but by the bandwidth of the source spectrum. The coherence properties play an essential role - to achieve high depth resolution, short coherence length or wide-band sources are required. Wide-band sources such as superluminescent diode (SLD), Kerr lens mode-locked laser and supercontinuum sources have been used for this purpose. Sub-micron resolutions are achievable using specialized ultra-wide band sources [17]. OCT gives higher dynamic range images than confocal microscopy because it takes advantage of the coherence properties of light and registers only correlated signals, and it amplifies the weak signal from the object arm by mixing it with the strong signal from the reference arm. A reflectivity below 10^-13 could in principle be measured with OCT. It is recognized that in comparison with confocal microscopy, OCT gives approximately double the penetration depth in highly scattering samples such as paint layers. Since OCTs are designed mainly for in vivo examinations of a highly sensitive organ - the eye, care must be taken to avoid any potential risk of contact. Hence it has been developed to perform high resolution scanning at comfortable working distances (typically a couple of centimetres), which is also an important requirement for safe scanning of valuable paintings.

Yang et al. [13] used a conventional A-scan based time-domain OCT to examine the cross-section of ancient jade. Targowski et al. [15] used a frequency-domain OCT to examine a variety of museum objects including a cross-section image of a section of a 19^th century painting showing a varnish layer over a paint layer. These systems based on A-scans cannot give real time images in the en-face orientation. In our preliminary studies [14], we examined a variety of paintings and paint samples using a different OCT imaging technology, operating en-face, which facilitates easy navigation around a painting since the real time en-face images acquired through OCT can be easily identified with features in the same orientation on a painting seen with the naked eye. These en-face OCT systems revealed not only varnish and paint layers but also the ground layer in some case. In this paper, we have extended our preliminary studies to include high resolution and dynamic range imaging of underdrawings which reveals not only the fine structures of the drawings but also the exact 3D position of the drawings. The fine structure of the drawings and the 3D images of the paint layers would enable conservators and art historians to determine, respectively, the type of drawing (e.g. solid or liquid based) and the layer on which the underdrawing was drawn.

2. The instruments

In this work we have used two different OCT systems that were assembled by the Applied Optics Group of the University of Kent. The optical configuration is similar to that presented in [18], using two single mode directional couplers with a superluminiscent diode as the source. Unlike conventional A-scan based time-domain OCT [16], en-face OCT systems [19] construct B-scans and C-scans from en-face (or T-scan) reflectivity profiles. This is similar to the procedure used in any confocal microscope, where the fast scanning is en-face and the depth scanning (focus change rather than a change in the reference arm length in the case of an OCT) is much slower (at the frame rate) [20]. The en-face scans provide an instant comparison to the familiar sight of a painting which is particularly convenient for navigating around a painting in a conservation examination. Features seen with the naked eye could easily be compared with features hidden in depth. Sequential and rapid switching between the en-face regime and the cross-section regime, specific for the en-face OCT systems [21] developed by us, represents a significant advantage in the non-invasive examination of paintings.

As shown in Fig. 1, in the en-face C-scan regime, the frame grabber is controlled by signals from the generators driving the X-scanner and the Y-scanner. One galvo is driven with a ramp at 500 Hz and the other galvo-scanner with a ramp at 2 Hz.

In this way, a C-scan image in the plane (x,y) is generated at constant depth. The next C-scan image at a new depth is then generated by moving the translation stage in the reference arm of the interferometer and repeating the (x,y) scan. Ideally, the depth interval between successive frames should be much smaller than the system resolution in depth and the depth change applied only after the entire C-scan image has been collected. However, in practice, to speed up the acquisition, the translation stage was moved continuously. In the images presented below, no other phase modulation was employed apart from that introduced by the X-galvanometer scanner. In Podoleanu et al. [18], we demonstrated the role played by the image size in balancing the effects of an external phase modulator and of the modulation produced by the transversal scanner. If the image is sufficiently large, then the distortions introduced by not using a phase modulator are insignificant.

In the en-face B-scan (cross-section) regime, the frame grabber is controlled by signals from the generator driving the X-scanner (or the Y-scanner) with a ramp at 500 Hz and the translation stage moving over the depth range required in 0.5 s. In this case, an OCT cross-section image is produced either in the plane (x,z) or (y,z).

The two systems operate at 850 and 1300 nm. The system at 1300 nm, has a low numerical aperture (NA) interface optics which gives a large field of view of 1 cm by 1 cm, but relatively low transverse resolution of 25 μm. The system at 850 nm, has a higher NA interface optics which gives a better transverse resolution of 15 μm. Both systems have typical working distances of 2 to 3 cm and depth resolutions of 18 to 20 μm (in air). The transversal resolution was measured by scanning a standard resolution target at the focal plane of the interface lens. With the transversal scanning switched off, the depth resolution was measured by placing a plane mirror in the object arm and recording the correlated signal strength as a function of the reference arm length. In order to image different objects, adjustable interface optics was designed. The 1300 nm system was used to survey the samples first, since it has a larger field of view and it was thought to have penetration for most pigments [4]. To make qualitative comparisons between the two wavelengths, images with the 850 nm system are also presented in some cases. We do not attempt to make quantitative comparisons in this paper, since it is only possible if images are acquired at exactly the same position at both wavelengths and preferably at the same resolution which would require a dual wavelength system.

 

Fig. 1. Optical coherence tomography system architecture for the examination of paintings. SLD = superluminescent diode, IMG = index matching gel

Download Full Size | PDF

3. Results

3.1 Varnish on glass substrate

A 1952 varnish sample on glass substrate was first measured with the 1300 nm OCT system by placing the sample ~2 cm from the probe head of the OCT (see Fig. 2). The refractive index of the varnish sample was measured to be 1.537±0.003 using an immersion method where tiny varnish samples immersed in liquids of calibrated refractive indices are observed under a microscope [22]. The varnish optical thickness was found to be 118 ± 5 μm which is ~77 μm in thickness for a refractive index of 1.537. The thickness of the varnish sample was measured independently by making a cross-section and viewing it under a microscope. This measurement found a varnish thickness of ~80–150 μm across a sample of 400 μm in size, which is consistent with the thickness measured by the OCT. The accuracy of the measurement under the microscope is limited by how the sample was taken and mounted, since it is difficult to cut a sample and mount it such that the exposed cross-section represents a cut exactly perpendicular to the painting surface. The thickness measured is an upper limit since any deviation from a straight cross-section would over-estimate the thickness.

 

Fig. 2. A 1300nm OCT image of a mastic varnish layer (on a piece of glass) dating from 1952; the varnish/air and varnish/glass interfaces are clearly delineated. The third (faint) interface is a ghost image or the result of multiple reflections between the varnish/air and varnish/glass interface. The vertical scale represents depth measured in air.

Download Full Size | PDF

3.2 Painted test samples

A series of sample boards to which oil paint layers had been applied was prepared. Films of dammar or mastic varnish were applied to either side of the sample leaving the centre region unvarnished (e.g. Fig. 3). In some of the samples the paint patches were applied with a brush, simulating a real painting, where the depth varies from point to point, while in others a film-spreader was used to give an ideal situation where the depth is uniform. The pigment mixtures and layer combinations reflect those found in real oil paintings. Ground, paint and varnish layers were painted onto six 7 cm by 8.5 cm Teflon (PTFE) boards. The surface of each Teflon board was sanded so that the ground layer would be able to bind to it. A traditional painting ground was made by the addition of chalk to rabbit skin glue. Teflon boards were either covered by spreading the ground to a thickness of 200 μm, or by brushing it on. Once dry, the ground was sealed by the application of two size layers so that the oil from the oil paint was not absorbed into the porous chalk ground. Diluted rabbit skin glue was used as the size. Two pigments in linseed oil were then painted onto each board.

Selected points on the sample board shown in Fig. 3(a) were imaged using the OCT systems. Figure 3 shows the potential of infrared OCT in obtaining cross-section information on paintings. The 1300 nm instrument was used to scan across the boundary between the yellow ochre and smalt paints on the sample board (Fig. 3(b)) which shows yellow ochre to be a highly scattering medium compared with smalt. The transparency of smalt to radiation at 1300 nm enables the OCT to see through it and measure the tomography of the ground layer beneath. For comparison, a cross-section image was obtained with the 850 nm system for the yellow ochre/smalt boundary on the varnished part of the panel (Fig. 3(c)) which shows the smalt layer to be more transparent than the yellow ochre layer similar to the images at 1300nm. Similarly, an 850nm image of a cross-section on the varnished smalt region shows clearly the varnish layer, smalt layer and the ground layer (Fig. 3(d)).

 

Fig. 3. (a) Sample board of yellow ochre and smalt on a ground layer of chalk and rabbit skin glue. One third of the sample has three coats of dammar varnish applied, and another one third of the painting has 3 coats of mastic varnish applied, leaving the central part of the sample unvarnished; (b) 1300nm OCT image of a cross-section across the yellow ochre/smalt boundary (scan in the middle of the panel; c-d) 850nm OCT images of cross-sections on the yellow ochre/smalt boundary (scan along the bottom line segment) and in the smalt area (scan along the top line segment). The vertical scale represents depth measured in air.

Download Full Size | PDF

3.3 Test paintings

A roughly 50-year-old test painting, which has half the aged yellow varnish removed, was used to test the technique on a real painting painted on canvas (see Fig. 4(a)). Two small patches of new mastic and dammar varnish were applied to the top left hand corner of the painting where the old varnish had been removed. At point B a layer of new mastic varnish had been applied after the old varnish was removed, and at point C no varnish was applied after the old varnish was removed. The 1300 nm OCT measurement of point B and C on the test painting confirmed that there was a thin layer of varnish at point B but no varnish at point C (Fig. 4(b), Fig. 4(c)). Figure 4(d) shows clearly that part of the painting is covered with two varnish layers, a thicker new varnish layer on top of an old varnish layer. Such information is useful for the understanding of the conservation history of a painting. Varnish thickness measurements on the test painting obtained using OCT systems working at 1300 nm and 850 nm (Fig. 4(e)) for the old varnish layer was found to be ~80 μm in the thicker regions, which is also consistent with the measurements made by cross-sectional analysis of samples taken from the painting.

 

Fig. 4. (a) A 50-year-old test painting: Point A is covered with the original varnish, which is now yellowed; Point B has the old varnish removed and new varnish applied; Point C is unvarnished; Point D is the boundary between regions where there is just one layer of old varnish and regions where a layer of new varnish was applied on top of the old varnish. (b-d) 1300nm OCT image of Point B, C and D. (e) 850nm OCT cross-section at Point A; The vertical scale represents depth measured in air.

Download Full Size | PDF

An 18th century painting, which has suffered severe damage where parts of the painting had lost the varnish or even the paint layer with bare canvas showing through, was used to test the technique on an ancient painting (see Fig. 5(a)). The 1300nm OCT scan of line segment A on the painting (Fig. 5(b)) shows that the upper half of the line is covered with a varnish of thickness ~90 μm and the lower half is devoid of varnish. The 1300 nm OCT measurement of line segment B on the 18th century painting shows that the varnish layer was not original but applied after the paint loss since a varnish layer is found on top of the region where the underlying canvas is seen (Fig. 5(c)).

 

Fig. 5. (a) An 18^th century panel painting; (b) a cross-section image of a scan along the top line-segment marked on the painting; (c) A cross-section image of along the lower line-segment on the painting. Both images were obtained with 1300 nm system.

Download Full Size | PDF

3.4 Imaging of underdrawings

Underdrawings are the preliminary sketches made by an artist before applying the paint layers. The study of underdrawings is particularly useful to the art historian for understanding painting techniques and for attributing works of art to specific artists. Near infrared imaging of paintings are routinely carried out in museums and galleries to detect underdrawings. Paint layers are relatively transparent in the near infrared compared with the visible. Since the 1970s, infrared vidicon tubes were used for routine examination of underdrawings in museums and galleries [4]. Recently, some galleries have acquired digital infrared cameras based on InGaAs chips. The InGaAs chips are nearly an order of magnitude more sensitive than the vidicon tubes, hence they are better at imaging underdrawings as can be seen in Fig. 6 (compare images in column b and column c). However, Fig. 6(d) shows that the 1300nm OCT is even better at imaging underdrawings than the InGaAs cameras. Since OCTs are only sensitive to coherent signals and they are depth selective, they can achieve better dynamic range compared with a direct imaging device which registers the sum of all reflected signals without discrimination. The OCT images shown in Fig. 6 are the addition of a number of en-face slices that contained underdrawings. Fig. 6 shows that the dynamic range and resolution of the OCT images of underdrawings surpass any conventional infrared images. The high dynamic range is because interferometers register only coherent signals hence only back-scattered light from the layer that matches (within the coherence length) the reference path length is registered. Back-scattered light from the other layers is automatically rejected.

 

Fig. 6. (a) Color images of two painted patch over underdrawings: the top patch has two layers of lead-tin-yellow paint over underdrawings drawn with a quill pen using an ink of bone black in gum; the bottom patch has a mixture of lead white, azurite, bone black painted over a black chalk underdrawing; (b) the corresponding near infrared Vidicon images; (c) the corresponding near infrared images taken with a InGaAs camera; (d) the corresponding 1300 nm OCT images taken at the depth of the underdrawings. The size of the images are ~ 1 cm by 1 cm.

Download Full Size | PDF

Figure 7 shows an example of a series of images taken in depth of a small region of a painted board, where the lower part of the panel is painted with a white translucent layer above some underdrawings drawn on top of a ground layer and the upper part is painted with an additional blue paint layer. A stack of 80 en-face C-scan images were collected while moving the reference stage at 50 μm/s which means 25μm in depth between the frames for a frame rate of 2 Hz. As images are taken deeper into the painting, underdrawings are revealed. A B-scan image in the white part of the panel (Fig. 7(f)) shows that the underdrawing is below one layer of paint and the B-scan image in the blue part of the panel (Fig. 7(g)) shows that the underdrawing is below two layers of paint. It should be appreciated that the comparisons shown in Fig. 7 has been made convenient owing to the capabilities of the en-face OCT systems. Such comparisons are not possible in real time using A-scan based OCT systems (conventional time-domain OCT or Fourier domain OCT).

 

Fig. 7. (a) Color image of a painted panel: the lower part is painted with an imprimatura (a translucent paint layer) on top of the underdrawing which is painted on a preparatory ground layer, the upper half has an additional paint layer above the imprimatura; (b) average of the top four en-face images collected with the 1300 nm system, i.e. 0–100 μm; (c) average of the next four en-face images (100–200 μm); (d) average of the next seven en-face images (200–375μm); (e) average of the next seven images (375–550μm); f) B-scan image in the white area of the panel showing the underdrawing below the first layer of paint (imprimatura); g) B-scan image in the blue area of the panel showing the underdrawing below two layers of paint.

Download Full Size | PDF

3.5 3D volume imaging

As described above, a stack of en-face images can be collected in a 1 cm by 1 cm area using the 1300 nm system. The stack of images can also be rendered in volume to visualise the painting in depth over an extended area to assist conservation examinations. Fig. 8 shows a volume rendering of the stack of images collected over the square area of the yellow ochre/smalt boundary of the test panel shown in Fig. 3(a). It shows that the transition from the painted area to the unpainted bare Teflon board, where the edge of the painted area is thicker than the rest of the painted area as would be expected from a hand painted panel where the paint would be thicker at the end of a brush-stroke.

 

Fig. 8. (2.5 MB movie) Volume rendering of the square area indicated on Fig. 3(a) of the test panel painted with smalt and yellow ochre. The scales are 1 mm in depth and 1 cm in the other two dimensions.

Download Full Size | PDF

4. Conclusions

Optical Coherence Tomography is a powerful non-invasive technique for probing into the depth of paintings, providing 3-D infrared image cubes that can show not only the structure of the paint and varnish layers but also reveal the underdrawings and their depth positions. The OCT images present better microscopic tomography of the surface of the varnish and paint layers than any other system currently employed in the examination of museum paintings. Its advantage over normal cross-section examination under a microscope being that it would be possible to make a cross-sectional examination across the surface of the paintings rather than only at those discrete points selected for sampling and it is non-invasive. It gives the best dynamic range and resolution images of underdrawings compared to any currently employed technique, since interferometers take advantage of the coherence properties of light. En-face OCT is particularly suited to the examination of paintings as it provides real time imaging in the en-face plane and the versatility to switch from cross-section B-scans to constant depth C-scans in real time.